The effectiveness of applying analytic methods to real data – in evaluating the interaction of a time-shared computer system and its users – is strongly supported by the author. The technique employed is to specify and measure various characteristics of such computer systems, to specify and measure relevant human characteristics, and, on the basis of these quantitative parameters, to predict the over-all performance of both as they “discuss” their problem in sequences of command and response. Performance metrics are defined for the evaluation of the computer-user interaction, such as the ratio of response time to the required processor time, which is shown to result in an accurate measure of how much attention the computer is paying to its “client.”
Effects of changing small details in an existing time-sharing system are simulated; the predictions made on the basis of these simulations are compared with the real data derived from actual measurements. Finally, the behavior of a broader class of systems is considered, through the use of a continuous Markov process model. Both the data and the system models will be most useful in evaluating proposed new systems.